From Data Burden to Breakthrough Engine: What Four Standards Leaders Told Us About AI-Ready Clinical Research
Standards get a bad rap. Ask most clinical teams and you’ll hear the same complaint: they slow things down, add paperwork, and exist mostly to make regulators happy. But on a recent Xtalks panel, four people who’ve spent their careers building and applying data standards made the opposite case: standards aren’t the burden. They’re the infrastructure that makes AI in clinical research actually work.
The panel, “From Data Burden to Breakthrough Engine: The Expanding Value of Standards,” brought together:
- Rebecca (Becky) Kush, President of Catalysis and founder/President Emeritus of CDISC, as moderator
- Ron Fitzmartin, Principal Consultant at Decision Analytics and former FDA informatics leader, who served as rapporteur for the ICH M11 Clinical Protocol Template working group
- Aaron Mann, CEO of the Clinical Research Data Sharing Alliance (CRDSA)
- Dipak Kalra, President of the European Institute for Innovation through Health Data and Professor of Health Informatics at UCL
- Sean Connolly, Chief Product and Technology Officer at Evinova
Here’s what stood out.
The protocol shouldn’t end its life as a Word document
Fitzmartin opened with the core problem: protocols get authored once, then reinterpreted from scratch by every team downstream. Clinical ops, data management, biostatistics, regulatory, the CROs, each group reads the same 300-page document and translates it into their own systems and their own understanding. That repeated interpretation is where delay, rework, and inconsistency creep in.
His fix is ICH M11, the guideline that reached Step 4 after an eight-year effort and defines a protocol template plus a technical specification for making protocol content computable. As he put it, the real shift is that the protocol should not be completed as a Word document and then reverse engineered into structured information. Instead, structure gets captured as the protocol is authored, mapped to CDISC’s Unified Study Definitions Model (USDM), and validated continuously rather than handed off after the fact.
The payoff isn’t abstract. Once a protocol is structured, checking whether an endpoint aligns with its objective, or whether the schedule of activities actually supports the endpoints, happens automatically and early, before it turns into an amendment. Fitzmartin was direct about what’s still missing: regulatory agencies need to mandate the data format, not just the template, the same way FDA once made SDTM and eCTD submissions required rather than optional.
AI doesn’t replace standards. It exposes why they matter more
Aaron Mann’s CRDSA started in 2021 out of the COVID-19 data-sharing push, when 20 pharma companies signed data-sharing agreements in ten weeks but actually sharing usable data proved far harder than the commitments suggested. CRDSA spent its first few years defining what “good shared data” looks like for secondary use, distinct from CDISC’s regulatory-submission focus.
The pivot came when Mann’s team asked whether AI could solve the interoperability problem from the reuse side instead of waiting for input data to get cleaner. His answer to “doesn’t AI just solve this?” was blunt: naive applications of generative AI to harmonize health records won’t produce a regulatory-grade audit trail, won’t guarantee privacy, and won’t be reproducible. Run the same harmonization job on the same data six months apart and you can get different results.
His conclusion inverts the usual AI narrative: it actually argues for more standards, more data models rather than less. The more consistent the incoming data and metadata, the more reliably an AI system behaves.
Europe’s patient summary project has a twist worth watching
Dipak Kalra walked through xSHARE, an EU project matching international patient summary standards (the ISO IPS) against clinical trial feasibility and protocol data elements. The team found meaningful overlap between what a patient summary needs and what trial teams look for when assessing eligibility or site feasibility, so they built “IPS Plus,” a pragmatic superset that folds in public health data needs without bloating the base standard.
Two forces are pushing this forward:
- Global convergence around the ISO IPS standard
- The European Health Data Space, a multi-billion-euro initiative requiring 27 countries to adopt a shared patient summary format
The detail that stood out most was patient-facing: under the regulation, patients can download their own fully computable, standardized health summary. Kalra’s vision is that patients could eventually use that data to self-screen against trial eligibility criteria before a clinician ever nominates them, flipping today’s model where patients often don’t know a relevant trial exists.
Kalra also pushed back on the idea that competing standards bodies fight over territory. CDISC, HL7, SNOMED and others sat in the same working group for xSHARE, and by his account, everybody went beyond their budgets in the project and felt an enthusiasm to work together, with the resulting resources released openly rather than guarded competitively.
Standards commitment and standards operation are different things
Sean Connolly’s segment focused on the gap between sponsors agreeing to a standard and actually running on it. His framework: there’s committing to standards, adopting them, and operationalizing them in real time, and most organizations, vendors included, stall between the first two.
He described three tiers vendors typically sit at: marketing commitment with no real investment, exporting to USDM only when data leaves the system while it lives in a proprietary model internally, and natively transacting in USDM (what Evinova aims for, with USDM Plus extensions where the standard has gaps).
On where AI actually helps, Connolly was careful not to oversell it: it’s not “plaster over holes,” but it does speed up the learning loop, helping teams translate an established standard into what their own practice looks like today. His practical advice to sponsors: don’t try to standardize everything at once. Break the path from final protocol to operational trial into critical-path activities, focus there first, and take incremental wins rather than waiting years for an end-to-end system.
The line worth remembering: if you get this right upstream, you’ve created an incredibly valuable learning loop for your organization. Without structured data on why a protocol got amended or a decision got made, that institutional knowledge lives in people’s heads instead of feeding back into the next study design.
The so-what
Strip away the acronyms (USDM, M11, IPS Plus, SDTM) and the panel converges on one idea: AI-enabled clinical research is only as good as the standardized, structured data underneath it. Standards aren’t the tax you pay before innovation starts. They’re what lets AI produce results that are auditable, reproducible, and actually usable at scale, whether that’s a computable protocol, a shared patient summary, or a harmonization platform.
The audience polls bore this out. At the start of the session, only a third of attendees called themselves standards “experts,” with 44% still learning. By the end, a fifth said their organization was already implementing computable protocols, but 40% said the question didn’t yet apply to them, suggesting the gap between standards ambition and standards operation is still wide across the industry.
This is exactly the gap Evinova’s Study Designer is built to close. It’s built on the same USDM backbone Connolly described, so cost estimates, benchmarks, and design scenarios connect directly to the structured protocol data, not a re-keyed copy of it. In practice: cost estimates in minutes at over 97% accuracy, $100M in savings delivered in year one and ~$365M by year four across programs using it, roughly a month faster study start-up, and 30% more performing sites with 66% more patients per site. It’s one concrete answer to the panel’s “operationalize, don’t just commit” challenge, standards adoption that shows up as real trial economics rather than a compliance checkbox.
Watch the on-demand webinar:
For a deeper discussion of how standards can improve interoperability, strengthen AI outputs, and accelerate research readiness, watch the full Xtalks webinar on demand.
Transform your clinical trials today
